Context-Aware and Statistical Association-Driven Entity Linking Model

Entity linking aims to map mentions in text to corresponding entities in a knowledge base, and as a fundamental task in the field of natural language processing, entity linking has been widely applied to many downstream tasks, such as knowledge graph complementation. In recent years, with the rapid development of large-scale pre-trained language models, language models based on pre-training of ultra-large-scale corpora have demonstrated superior performance on entity linking tasks. However, most of the current entity linking models have the following problems: they cannot adequately capture the deep semantic relationship between mentions and candidate entities and they cannot accurately recognize the relationship between mentions and candidate entities in long-tailed entity scenarios. To solve the above problems, this paper proposes a context-aware and statistical association-driven entity linking model. The model uses Bi-GRU to further capture the deep sequence dependencies of mention contexts on the basis of encoding mentions and mention contexts in text using BERT. The candidate entities are also modeled using graph attention network and then further processed using Bi-GRU. In order to improve the model's ability to recognize long-tailed entity scenarios, this paper introduces the PMI value and combines it with the cosine similarity to rank the set of candidate entities to complete the entity linking. Experimental results show that the proposed model can effectively improve the performance of entity linking.